Un-expand $Id$ in a git-svn repository, with some special powers

fix-expanded-svn-id.sh

#!/bin/sh

# The problem:
#
# You have a clean git-svn working copy and, for example, rsync the files from
# the production server into it in order to find out whether some files have
# been changed directly on the server, without committing.
#
# However, the files on the production server possibly have $Id$ expanded, and
# now you're seeing a _lot_ of changed files which contain no differences except
# for the $Id$ expansion.
#
# So you're using sed magic to replace expanded $Id$s by unexpanded ones, but
# now you find out that sed will ensure that files have a newline at EOF. Which
# is basically a good thing, but if your repository contains files without these
# newlines at EOF, sed will add the newlines and git will consider them changed.
#
# You _could_ "fix" these files and commit the EOF changes, but if you can't or
# don't want to, you can use this script.
#
# It will unexpand $Id$ (could be extended to unexpand other keywords as well)
# and then do a "git diff" on the file to check whether its only change is the
# trailing newline. If so, it will run a "git checkout" on it in order to reset
# it to the version in your repository.

git status -s | sed -n -e 's/^ M \(.*\)$/\1/p' | while read -r file; do
	sed -i '' -e 's/\$Id: [^$]*\$/$Id$/g' "$file" && \
	[ "$(git diff -U0 "$file" | awk '
		NR == 6 { a = substr($0, 2); };
		NR == 7 { nonl = (substr($0, 1, 1) == "\\"); };
		NR == 8 { b = substr($0, 2); };
		END { if ((NR == 8) && (a == b) && nonl) { print "nonl"; } };
	')" = 'nonl' ] && echo "$file"
done | xargs git checkout