Merge multiple local Git repositories.

git-merge-repos.sh

#!/usr/bin/env bash

# A shell script that merges local Git repositories into a new repository. The
# script accepts multiple directories and/or glob patterns as arguments. Each
# local repository will be moved into a subdirectory, named by the user via
# stdin.
#
# Dependencies: bash 4+, git and git-filter-repo.

GIT_REPOS=( "$@" )
if [ ${#GIT_REPOS[@]} -eq 0 ] ; then
  echo "Please specify local Git repositories to merge."
  exit 1
fi

declare -a ABS_GIT_REPOS
for FOLDER in "${GIT_REPOS[@]}" ; do
  ABS_PATH="$(readlink -f "$FOLDER")"
  if [ ! -d "$FOLDER/.git" ] ; then
    echo "Skipped $ABS_PATH (no Git repository found)"
    continue
  fi
  ABS_GIT_REPOS+=("$ABS_PATH")
done

NEW_REPO="$(mktemp --directory --suffix="-merge-repo")"
git init "$NEW_REPO"

(
  cd "$NEW_REPO" || exit
  git commit --allow-empty --message "Initial commit"

  for FOLDER in "${ABS_GIT_REPOS[@]}" ; do
    REMOTE_NAME="$(echo -n "$FOLDER" | sha256sum | cut -c-32)"
    REMOTE_HEAD="$REMOTE_NAME/HEAD"

    git remote add --fetch --tags "$REMOTE_NAME" "$FOLDER"
    git remote set-head --auto "$REMOTE_NAME" 2>/dev/null
    if [ "$?" -eq 1 ] ; then
      echo "Skipped $FOLDER (no HEAD found)"
      continue
    fi

    read -r -p "Path to move files from $FOLDER [$REMOTE_NAME]: " NEW_FOLDER
    NEW_FOLDER="${NEW_FOLDER:-REMOTE_NAME}"

    MESSAGE="commit.message = b'$NEW_FOLDER: ' + commit.message; return commit"
    git-filter-repo --force --refs "$REMOTE_NAME" --prune-empty always \
      --commit-callback "$MESSAGE" --to-subdirectory-filter "$NEW_FOLDER"

    git merge --allow-unrelated-histories --no-edit "$REMOTE_HEAD"
  done

  printf -v LOG_FMT '%s' \
    "%at pick %h %s %x07@@@@@@@@@@" \
    " exec GIT_COMMITTER_NAME='%cn' GIT_COMMITTER_EMAIL='%ce'" \
    " GIT_COMMITTER_DATE='%cd' git commit --amend --no-edit"

  # The pipeline below acts on the git-log output via the following steps:
  #
  #  * sort by commit author date obtained as Unix timestamps (this is
  #    a personal preference and can be changed as desired);
  #  * substitute the BEL characters to insert the `exec` git-rebase commands
  #    directly below their respective `pick`s. Inserting the timestamp before
  #    the `exec` screws up sorting if there are commits done at the same time;
  #  * remove the last `exec` command because it will fail on an empty
  #    commit. This cannot be done on the previous `sed` pass because it would
  #    delete the line before being split, and so git-rebase would complain
  #    about missing commits;
  #  * lastly, remove the first twelve characters corresponding to the
  #    timestamps and the repeated @ characters so that the git-rebase todo
  #    syntax is correct.
  #
  # A temporary file is needed because Git may fail due to an excessively long
  # argument list.

  REBASE_COMMANDS="$(mktemp --suffix="-rebase")"
  git --no-pager log --no-merges --pretty="$LOG_FMT" \
    | sort -nk1 \
    | sed 's/\a/\n/' \
    | sed '$d' \
    | cut -c12- \
    > "$REBASE_COMMANDS"

  # Now we pass the organized git-rebase todo via the GIT_SEQUENCE_EDITOR
  # environment variable, which redirects it to the correct file.

  GIT_SEQUENCE_EDITOR="cat $REBASE_COMMANDS >" git rebase -i --root

  echo "Finished creation of merged repository at $NEW_REPO"
)