git_merge_fix_false_deletions.sh

#!/bin/bash

# This script is meant to help with updating old PRs.
# Mainly any OpenSCAD branch which has not been updated since 2022-02-06
# It is intended to be run on a git repo in which a merge has already been started (merging master into the current, old branch),
# and has conflicts which show as "deleted by them:", or "(modified/deleted)".
#
# Due to how git DOES NOT track file moves/renames, in addition to how it determines file "similarity" (by # of exactly matched lines in a file),
# it makes it very difficult to merge master into branches created before 2 significant events in the OpenSCAD repo:
#   1) A large code style reformatting (PR #4095 on 2022-02-06), and
#   2) A large restructuring of the source tree (PRs #4159 on 2022-03-06, #4161 on 2022-03-12)
#
# When attempting to merge master into such PR branches, the files are wrongly detected by Git as (modified/deleted).
# That is, modified locally (fine), but deleted in the remote master (WRONG).

#### THE PROBLEM ####
# Git doesn't track the renaming of files.
# Instead it attempts to detect them after the fact, through countless semi-functional hacks.
# This detection fails in our particular situation, making a proper 3-way merge extremely difficult.
#
#### THE SOLUTION ####
# This script goes through the list of conflicting files, determines which ones were presumed "deleted by them",
# properly follows the renames to determine the final destination path, and then rewrites the git index
# in such a way that "git mergetool" will detect them as "both modified".
#
#### CAVEATS and SHORTCOMINGS ####
# - This script doesn't work the other way around, for files renamed/moved in the current branch.
# - Only designed for merge, not sure about rebase
# - There is not currently any error checking or backup mechanism built into the script, so please USE WITH CAUTION

if [ "$#" -ne 2 ]; then
    echo "Usage: git_merge_fix_false_deletions.sh <remote> <branch>"
    echo "e.g.   git_merge_fix_false_deletions.sh origin master"
    echo "Please read over comments in this script before running."
    exit 1
fi

REMOTE=$1
BRANCH=$2
DIFFBASE="${REMOTE}/${BRANCH}"

ANCESTOR=$(git merge-base HEAD "$DIFFBASE")

echo -n "Using common ancestor:"
git -P show -s --pretty='format:%C(auto)%H (%an, %as) %s%n    %b%n' $ANCESTOR

FILES=$(git diff --name-only --diff-filter=U) # get a clean list of just the files which are (U)nmerged, (because they have conflicts)

echo
echo "Files modified between current branch and $DIFFBASE:"
while IFS= read -r FILE; do
  SRC="${FILE}"
  echo -n "  ${SRC}"
  DELETED_IN=$(git -P log $DIFFBASE -M --follow --diff-filter=D --pretty=format:"%H" -- $SRC)
  while [[ $DELETED_IN ]]; do
    #echo -n "deleted in "
    #git -P show -s --pretty='format:%C(auto)%H (%an, %as) %s%n' $DELETED_IN
    STAT_LINE=$(git -P show --name-status $DELETED_IN | grep -F $SRC)
    REGEX="^[[:space:]]*R[0-9]+[[:space:]]+([^[:space:]]+)[[:space:]]+([^[:space:]]+)[[:space:]]*$"
    if [[ $STAT_LINE =~ $REGEX ]]; then
      SRC="${BASH_REMATCH[1]}"
      DST="${BASH_REMATCH[2]}"
      echo -n "  ==>  ${DST}"
    else
      echo -n "  ==>  (actually deleted?)"
      DST=""
      break
    fi
    SRC="${DST}"
    DELETED_IN=$(git -P log $DIFFBASE -M --follow --diff-filter=D --pretty=format:"%H" -- $SRC)
  done
  echo
  if [[ $DST ]]; then
    mkdir -p $(dirname $DST)
    # have to clear index entry before updating, with 0 file mode, and all 0's SHA1
    echo "0 0000000000000000000000000000000000000000	${DST}" > .index-info.tmp
    git ls-files --stage ${FILE} | sed "s^${FILE}^${DST}^g" >> .index-info.tmp
    git ls-files --stage ${DST} | sed "s^ 0\t^ 3\t^g" >> .index-info.tmp
    git update-index --index-info < .index-info.tmp # Repair the conflict
    git rm ${FILE} # remove local old-named file
    DST="" # clear DST for next file
  fi
done <<< "$FILES"
rm .index-info.tmp
echo