Well, I finally got off my duff and did something with all the changes
I've been accumulating: I put my patch tree up for ftp.  It's available
from collatz.mcrcim.mcgill.edu in /ftp/mouse under the name
netbsd-patches.tar.gz; if you want individual files, there's a
netbsd-patches directory that contains the same stuff unpacked.  (Yeah,
yeah, I can hear the snickers already...just in case someone out there
is interested in some of these, okay?)

The -current patch tree contains a whole bunch of stuff.  On this list,
they're marked with the affected architecture in [ ], or [mi] for
machine-independent stuff.  I've sent in some of these via send-pr;
I've even seen a few of them fixed according to source-changes message,
but I haven't brought my base tree up-to-the-minute with the supped
tree in a couple of weeks, so the patches are still in my tree.  A few
more are things that have been turned down for the main tree, but I
still want, so they're in my patch tree.

	- the kernel config I use (and a patch to Makefile.sparc to
	    help with debugging kernels) [sparc]
	- a bunch of changes to the console terminal emulator [sparc]:
		- if RCONS_WONB in kernel config, don't come up in
		    reverse video
		- if RCONS_SMALL in kernel config, use a smaller font
		    (specifically, the "fixed" font from X)
		- if RCONS_FULLSCREEN in kernel config, use the whole
		    screen, regardless of what the PROM settings for
		    screen-#columns and screen-#rows are
		- never strip the high bit from output characters, plus
		    tweaks to make high-half characters work (char
		    becomes unsigned char several places)
		- add (most of) the Latin-1 high-half printable
		    characters to gallant.r.19, in case !RCONS_SMALL
		    (RCONS_SMALL (=fixed) already has these characters)
	- a patch to add tickadj to kernfs [mi]
	- a patch that makes set-id bits take effect in exec() even
	    when the process is being traced, provided the tracing
	    process is root [mi]
	- my md5-based libcrypt [mi]
	- a patch to get a warning when cc -Wformat -pedantic sees the
	    q format modifier [mi]
	- an attempt to reduce spurious DMAWAIT1 panics [sparc]
	- a patch someone sent me to si.c, hoping to help with Exabyte
	    trouble I was having (haven't had a chance to test) [sun3]
	- patches to give meaning to symlink permission bits: r permits
	    readlink()ing the link, x permits following it in namei()
	    (these patches definitely count as experimental) [mi]
	- a trivial fix to the mount() prototype in <sys/mount.h> [mi]
	- the fix for vm/vm_swap.c versus physio() [mi]
	- that patch for vnode_pager_setsize() to make it handle
	    partial pages better; also giving access to vpagerdebug
	    (via sysctl) [mi]
	- make fsck ignore clean bits when a filesystem is explicitly
	    named on the command line [mi]
	- make memory XDR streams (ie, xdrmem_create()d streams) work
	    with unaligned buffers, even on architectures that don't
	    like unaligned accesses [mi]
	- make tsort optionally quiet about cycles, plus make
	    <bsd.lib.mk> use this option [mi]

I'll be glad to correspond with anyone about any of the above.  (Note:
the README file is slightly out of date, especially with respect to the
SPARC rcons stuff.)

					der Mouse

			    mouse@collatz.mcrcim.mcgill.edu