Re: Identify Commit ID from an Extracted Source Snapshot

From: Jonathan Nieder <jrnieder@gmail.com>
Date: Feb 16, 2012, 23:53 UTC
Message-ID: <20120216235240.GA20779@burratino>
In-Reply-To: <4F3D8A7C.2020400@elegosoft.com>

Michael Schubert wrote:

> If your question is more like "how do I tell git to find out where
> this old code fits in my history and eventually place it there",
> the answer is: you cannot do it. No VCS will do this and especially
> not Git.

Wouldn't it be possible to add the tags you want by walking through the commit log to find a matching commit for each tarball?

For example:

	# Usage: "GIT_DIR=<repository> tag-tars <tarballs>"
	# Arguments should be tarballs containing releases in
	# reverse-chronological order.
	# Should be run in an empty directory, which will be
	# used as a workspace.

	# save stdin
	exec 3<&0

	GIT_DIR=$(git rev-parse --resolve-git-dir "$GIT_DIR") || exit 1
	GIT_INDEX_FILE=$GIT_DIR/index.tag-tars
	export GIT_INDEX_FILE

	if test -n "$(git ls-files -c -o | head -1)"
	then
		echo >&2 'fatal: I need an empty directory to work with'
		exit 1
	fi

	for tar
	do
		# empty workspace
		git ls-files | xargs rm -f --
		rm -f "$GIT_INDEX_FILE"

		# get tree name for tarball
		tar --strip-components=1 -xf "$tar"
		git ls-files -o | git update-index --add --stdin
		tree=$(git write-tree)

		# tag the first commit found matching that tree, if any
		git rev-list master |
		while read cmit
		do
			if git diff-tree --quiet $cmit $tree
			then
				git tag -a ${tar%%.*} $cmit <&3
				break
			fi
		done
	done

Variation using --numstat and a path filter to find the closest commit ignoring some files instead of an exact match left as an exercise to the reader.

Hope that helps, Jonathan