Making custom git-remove-tree command

I'm doing a script named git-remove-tree which removes the working tree known to Git. It doesn't touch untracked files; it only deletes directories if they are empty. The script seems to work, but because I'm not very good at Git plumbing and there can be some corner cases which I don't know about, I'd appreciate if more experienced users would have a look. Is the following script safe?

#!/bin/sh # git-remove-tree

is_wt=$(git rev-parse --is-inside-work-tree)
if [ "$is_wt" = false ]; then
	echo "You must run this inside a working tree."
	exit 1
elif [ -z "$is_wt" ]; then
	exit 1
fi

top=$(git rev-parse --show-toplevel)
cd "$top" || {
	echo "Can't go to top-level directory $top"
	exit 1
}

git ls-tree --name-only -r -z HEAD | xargs -0r -- sh -c '
for f in "$@"; do
	rm -f -- "$f"
	d=$(dirname -- "$f")
	[ "$d" = . ] || rmdir -p -- "$d" 2>/dev/null
done' dollar0_argument

echo 'Use "git reset --hard HEAD" to populate the working tree again.'

Teemu Likonen schrieb:

I think you should be able to reduce all of this to just

git read-tree --index-output=/tmp/empty.index -m -u \
	4b825dc642cb6eb9a060e54bf8d69288fbee4904 # empty tree
rm -f /tmp/empty.index

(with the usual caveats about temporary files).

-- Hannes

* 2010-02-04 10:55 (+0100), Johannes Sixt wrote:

Thanks, it works. It's also _much_ faster than my "git ls-tree" with shell loop thing. It prints ugly error, though:

    fatal: unable to write new index file

Also, it doesn't leave any index output files around.

Teemu Likonen schrieb:

Then put the temporary index into your git-dir:

   gitdir=$(git rev-parse --git-dir)
   git read-tree --index-ouput="$gitdir/tmpindex" ...
   rm -f "$gitdir/tmpindex"

The purpose of this temporary index is only that your real index is not cleared. If you don't mind that your index is purged as well, then you can just drop the --index-output parameter.

-- Hannes