After a number of setbacks over what's really only a couple of weeks, but
which feels much longer, I've finally succeeded in building release sets
of -current (as of about 7 November) all the way through with no errors.

I started back when the cache-flush bug was squashed.  The first problem
was that some files were still out of sync.  Updating frequently eventually
saw everything set straight.  Next, my 'big' SS5-110/256MB was going
comatose or crashing during builds, whether running multiple jobs (-j 4)
or only one.  This turned out to be interesting.

With the cache-flush bug fixed, my SS5 (which had never panicked before,
ever) was now running up against the "out of space in kmem_map" panic
problem.  While annoying, the workaround is straightforward and indicates
that NetBSD/sparc now behaves consistently across multiple CPU variants
with large quantities of system RAM.

So, I doubled NKMEMPAGES_MAX (to 3072), rebuilt the kernel yet again and
started over (I think I re-synced to head in the meantime).  I haven't
suffered any more panics on this machine since then.

The last problem was that DESTDIR was dirty.  My previous build attempts,
both successful and unsuccessful had left files lying around that were
not being cleaned out (especially when restarting a failed build with
'-u').  Some were over a month old.  The hangup was in the object files
for Xnest.  So, I wiped just that directory out and started over yet
again.  Restarting the build finished correctly.

After installing this distribution, and noting the lessons learned, I
sync'ed with HEAD one more time, wiped out the existing destination
and release directories, and started the build one more time, building
first tools and my custom kernel, rebooting and building distribution,
installing the distribution and then building release.  I pushed it
harder (-j 6) at every phase and it complained not at all.

Everything built top to bottom with no errors and runs very well.  I
should note that I built everything with "-mcpu=v8" as this seems the
most reasonable on the MB86904 microSPARC-II CPU.  Code size seems
smaller than '-mcpu=supersparc' and it feels just a bit faster, too.

Thanks to everyone who works on NetBSD/sparc.

-- 
John D. Baker, KN5UKS                    NetBSD     Darwin/MacOS X
jdbaker(at)mylinuxisp(dot)com                 OpenBSD            FreeBSD
BSD -- It just sits there and _works_!