Earlier, I upgraded to cvs-HEAD to get the reintegration fix patch,
and ended up with buggy rpc2 code and all my venii crashed :-(.  I
backed off from that and now all is ok.

In the middle of all this, I'm also debugging significant surgery to
the PF_KEY code on my laptop.  Bugs in my code caused IPsec to be
unstable, and thus I didn't run racoon (IKE daemon), and venus
couldn't talk to the coda server.  I had been editing my software
development log file, but was losing due to the emacs/NetBSD/SIGIO
problem, so I stopped and kept a local copy.

Then, I wanted to apply a patch to du I had developed for FreeBSD (to
not report sizes of files marked 'nodump').  Without thinking, I
grabbed the patch out of /coda/home/gdt/PATCHES/FREEBSD, edited it
into the NetBSD source, and ran 'cvs diff' redirecting to
/coda/home/gdt/PATCHES/NETBSD/du-nodump.   I did this all forgetting
that I had no connectivity to the server.

Later, I got IPsec stable enough, rebooted, started up racoon and
venus, and my changes got reintegrated correctly as soon as I got
tokens.  I know this _shouldn't_ be exciting, but it was, particularly
given my recent adventures.

On the subject of emacs, compiling emacs21 with

#define BROKEN_SIGIO

added to
/usr/pkgsrc/editors/emacs21/work/emacs-21.2/src/s/netbsd.h

seems to result in satisfactory operation.  Thanks to Brett for the
hint.  So, it seems that the coda kernel code in NetBSD 1.6 doesn't
handle signals well.