I'm running NetBSD/i386 -current as of Dec 1, and after a 'make snapshot'
with Dec 18th's sources, I was in single-user mode, copying files over, and
noticed some lossage on nearly full partitions.

My / looks like this:

Filesystem  1K-blocks     Used    Avail Capacity  Mounted on
/dev/wd0a       18259    13254     4092    76%    /


(Yes, it's tiny, but I like it that way)

I have softdep enabled on that partition, and when doing either
"tar --unlink -zxf base.tgz -C /" or "pax -rw -pe . /sbin/" I would get part
way through the transfer, and then get 'disk full' messages.  Obviously this
isn't desirable, and if I copy over the files in smaller batches (say 2-3
items from sbin at a time), then everything worked.  It APPEARS that with
softdep/trickle-sync, it's much easier to temporarily run out of disk space.
When pax would start giving me 'disk full' errors, I'd do a 'df' and notice
that / was most certainly full.  Running 'sync' (or waiting a few moments)
and running 'df' again would show 4M free again.

This doesn't seem to be the proper operation...has anyone else seen this?
It seems like if a partition is at capacity, the filesystem should clean
things up and try again before failing.

    ScottE