delete tempdir

	'git subtree rejoin' option to do the same as --rejoin, eg. after a
	  rebase

	--prefix doesn't force the subtree correctly in merge/pull:
	"-s subtree" should be given an explicit subtree option?
		There doesn't seem to be a way to do this.  We'd have to
		patch git-merge-subtree.  Ugh.
		(but we could avoid this problem by generating squashes with
		exactly the right subtree structure, rather than using
		subtree merge...)

	add a 'push' subcommand to parallel 'pull'
	
	add a 'log' subcommand to see what's new in a subtree?

	add to-submodule and from-submodule commands

	automated tests for --squash stuff

	test.sh fails in msysgit?
		sort error - see Thell's email

	"add" command non-obviously requires a commitid; would be easier if
		it had a "pull" sort of mode instead

	"pull" and "merge" commands should fail if you've never merged
		that --prefix before
		
	docs should provide an example of "add"
	
	note that the initial split doesn't *have* to have a commitid
		specified... that's just an optimization

	if you try to add (or maybe merge?) with an invalid commitid, you
		get a misleading "prefix must end with /" message from
		one of the other git tools that git-subtree calls.  Should
		detect this situation and print the *real* problem.
	
	In fact, the prefix should *not* end with slash, and we
		should detect (and fix) it if it does.  Otherwise the
		log message looks weird.
		
	totally weird behavior in 'git subtree add' if --prefix matches 
		a branch name

	"pull --squash" should do fetch-synthesize-merge, but instead just
		does "pull" directly, which doesn't work at all.

	make a 'force-update' that does what 'add' does even if the subtree
		already exists.  That way we can help people who imported
		subtrees "incorrectly" (eg. by just copying in the files) in
		the past.

	guess --prefix automatically if possible based on pwd

	make a 'git subtree grafts' that automatically expands --squash'd
		commits so you can see the full history if you want it.