GitHub source zipball reproduction attempt

zip.bash

#!/usr/bin/env bash

# Example:
#
# 1. git clone [email protected]:elm/core.git
# 2. cd core
# 3. ./zip.bash 1.0.5 out.zip elm/core
# 4. Compare outputs of:
#    - `unzip -l out.zip` and `unzip -l github-zipball.zip`
#    - `wc -c out.zip` and `wc -c github-zipball.zip`

# Exit on errors.
set -e

# Read command line arguments.
tag="${1:?'You must provide a tag as the first argument.'}"
new_zip="${2:?'You must provide the output zip file as the second argument.'}"
repo="${3:?'You must provide the repo (user/reponame) as the third argument.'}"

# Turn the tag into a full commit hash.
commit="$(git rev-list -n 1 "$tag")"

# Get short hash ID of the annotated tag.
directory_suffix="$(git rev-parse --short "$tag")"

# Switch to the given commit.
git switch --detach "$commit"

# Format commit date for use with `touch` below.
date="$(TZ=UTC0 git show --quiet --date='format-local:%Y-%m-%dT%H:%M:%SZ' --format="%cd")"

# Create the directory to zip. Remove one created from previous runs first.
dir="${repo/\//-}-$directory_suffix"
rm -rf "$dir"
mkdir "$dir"

# Keep track of files and directories to put in the zip.
files=()
dirs=("$dir/")

# Go through all files and folders checked into git.
for original in $(git ls-tree -r -t --full-tree --name-only "$commit"); do
  f="$dir/$original"

  if test -d "$original"; then
    # Give a trailing slash to directories. This is needed for sorting later.
    dirs+=("$f/")
    # Create directory.
    mkdir -p "$f"
  else
    files+=("$f")
    # Copy file to the zip directory.
    cp "$original" "$f"
    # Set mtime to the commit date.
    touch -d "$date" "$f"
  fi
done

for f in "${dirs[@]}"; do
  # Set mtime to the commit date for all folders.
  # This has to be done after all files are copied, since writing a file
  # updates the mtime of folders.
  touch -d "$date" "$f"
done

# Remove zip created from previous runs of this script, just in case.
rm -f "$new_zip"

# Create the new zip.
# -z creates a comment for the whole zip. The contents for the comment are passed on stdin via echo.
#    GitHub seems to use the commit hash as comment.
# -9 is max compression. However, GitHub’s zips are even smaller. Not sure how they do that.
# The sort order is important – it’s preserved in the zip file.
all=("${dirs[@]}" "${files[@]}")
echo "$commit" | zip -z9 "$new_zip" $(IFS=$'\n'; echo "${all[*]}" | sort) >/dev/null

I got a tip that https://diffoscope.org/ is kind of made for this problem. It shows that for example file permissions differ.