These should not be interpreted as things I am going to do.
These are things that I thought should be done at the time
I made these notes.

------

systems without fchown/fchmod?

make slist storage completely dynamic.  Rip out more static buffer
sizes (use str3save())

perhaps there should be some way to have a script but indicate that one
doesn't want the filenames on stdin.

extend test suite:
	exceptions, patterns
	different cmp file, dest file
	single file entry
	single link entry
	followlink?
	normal case of one machine to another with no presubbing.
	
global exceptions should be stored more sensibly -- adding them to
each slist entry is wasteful.

does followlink actually work?  doesn't look it in the dofinds code...
hm, maybe this happens in the slist reading code.  see the resolve
stuff.

remove -r?  (can be intuited from the slists line it gets which is of
the form rrootdir/stats/slist.librarian_name)

-S special files

option not to track over setuid files.

get_head shouldn't read 1 byte at a time.  that would mean get_head,
passybytes and skipbytes would have to share the buffering scheme.

bootstrap

	should be able to ask a hsot for all slists it exports this
	means a distinction between exported slists and imported
	ones.  Perhaps they should go into separate directories
	(slists.export, slists.import)

to handle 14 char filenames and use stats/hostname/subname instead of
stats/subname.hostname?

-R
	to do a -R, we have to do a special version of -w and a -n.
	for each slist entry {
		presub the fromfile with the cmpfile
		(if cmpfile != destfile, then what?)
		run the -w on the cmpfile, applying exceptions
		in the usual way to a statfile.local
	}
	get the remote statfile with a notreally, presub, apply
	exceptions, and print to a statfile.remote.transformed.

	now compare the local and remote.transformed statfiles.
	anything that is in local and not in remote.transformed is
	a candidate for removal.  print it.
