A `git filter-branch` script for extracting arbitrary files in a repo.

git_file_filter.sh

#!/bin/bash
#
# usage: git filter-branch --force --prune-empty --index-filter "$(cat git_file_filter.sh)" --tag-name-filter cat -- --all
#

shopt -s globstar

whitelist=(
# files and globs go here
)

function whitelisted()
{
	for path in "${whitelist[@]}"; do
		if [[ "$1" == "$path" ]]; then
			return 0
		fi
	done

	return 1
}

delete=()

for file in $(git ls-files); do
	if ! whitelisted "$file"; then
		delete+=("$file")
	fi
done

git rm --cached --ignore-unmatch "${delete[@]}"

Apparently foo/{,**/}*.rb globs do not work in this script. Had to explicitly expand them.

In the year 2021 git filter-branch is kind of discouraged due to performance and gotchas. Use git filter-repo or BFG. I successfully used:

git filter-repo --force --path PATH1 --path PATH2 ... --tag-rename '':'old-repo-'