#!/usr/bin/env bash

# file      : bpkg-rep/publish.in
# copyright : Copyright (c) 2014-2018 Code Synthesis Ltd
# license   : MIT; see accompanying LICENSE file

# Update (bpkg-rep-create(1)) and publish (rsync(1)) an archive-based
# repository.
#
# Pull a pre-cloned (read-only) git repository with the contents of an
# archive-based bpkg repository. Bail out if nothing changed from the the
# previous run. Otherwise, regenerate the repository meta-data by running
# bpkg-rep-create(1) on each section of the repository and, optionally,
# synchronize it to one or more destinations with rsync.
#
# The repository contents are expected to be in the <dir>/1/ subdirectory. The
# script saves the last successfully published commit in the <dir>.publish
# file.
#
# --destination|-d <host>:<dir>
#
#   Remote host and directory to rsync the repository to. Note that the
#   directory should include the 1/ component and any sub-directories that may
#   follow. In other words, the rsync command will be in the form:
#
#   rsync ... <dir>/1/ <host>:<dir>/
#
#   Repeat this option to specify multiple destinations. In this case, the
#   destinations are synced in the order specified with the first failure
#   terminating the process (so if you have a "primary" destination and a
#   "mirror", you probably want to specify the former first).
#
# --timeout <seconds>
#
#   Git and rsync operation timeout. Specifically, the operation will be
#   aborted if there is no network activity for the specified time. Default is
#   60 seconds. Note that currently the git timeout is only supported for the
#   http(s) transport.
#
# --lock-timeout <seconds>
#
#   The repository lock timeout. Fail if another instance of the script does
#   not release the repository in the specified time. The default is 0 (do
#   not wait).
#
#   Note that you will most likely want to specify a non-zero timeout for cron
#   jobs that may potentially overlap.
#
# --log-dir <dir>
#
#   Directory to create the temporary log files in. If unspecified, the stderr
#   is not redirected and no log is created by default.
#
#   The log is dumped to stderr in case of an error or at the end of execution
#   unless in the quiet mode and is then deleted.
#
# --quiet
#
#   Run quiet. Specifically, don't dump the log to stderr on exit with zero
#   status.
#
# --config <path>
#
#   The configuration file containing a bash fragment. Repeat this option to
#   specify multiple configurations that will be sourced in the order
#   specified.
#
# --bpkg <path>
#
#   The package manager program to be used for the repository update. This
#   should be the path to the bpkg executable.
#
usage="usage: $0 [<options>] <dir> [<rep-create-options>] [-- <rsync-options>]"

trap "{ exit 1; }" ERR
set -o errtrace # Trap ERR in functions.

@import bpkg-rep/utility@

# The script own options.
#
repo_ver=1
destinations=()
timeout=60
lock_timeout=0
log_dir=
quiet=
configurations=()
bpkg=

while [ $# -gt 0 ]; do
  case $1 in
    --destination|-d)
      shift
      destinations+=("${1%/}")
      shift || true
      ;;
    --timeout)
      shift
      timeout="$1"
      shift || true
      ;;
    --lock-timeout)
      shift
      lock_timeout="$1"
      shift || true
      ;;
    --log-dir)
      shift
      log_dir="${1%/}"
      shift || true
      ;;
    --quiet)
      shift
      quiet=true
      ;;
    --config)
      shift
      configurations+=("$1")
      shift || true
      ;;
    --bpkg)
      shift
      bpkg="$1"
      shift || true
      ;;
    --)
      shift
      break
      ;;
    *)
      break
      ;;
  esac
done

# The repository directory.
#
repo_dir="${1%/}"
shift || true

# bpkg-rep-create options.
#
rep_create_options=()

while [ $# -gt 0 ]; do
  case $1 in
    --)
      shift
      break
      ;;
    *)
      rep_create_options+=("$1")
      shift
      ;;
  esac
done

# rsync options.
#
rsync_options=()

while [ $# -gt 0 ]; do
  rsync_options+=("$1")
  shift
done

# Validate options and arguments.
#
if [ -z "$repo_dir" ]; then
  error "$usage"
fi

if [ ! -d "$repo_dir" ]; then
  error "'$repo_dir' does not exist or is not a directory"
fi

# If the log directory is specified then redirect stderr to the log file and
# setup the trap that dumps it on exit, if required.
#
if [ -n "$log_dir" ]; then

  if [ ! -d "$log_dir" ]; then
    error "'$log_dir' does not exist or is not a directory"
  fi

  # Create the log file.
  #
  log="$(mktemp "$log_dir/$(basename "$repo_dir").XXXXXXXXXX")"

  # Save the stderr file descriptor so we can dump the log into it on exit, if
  # required. Then redirect it to the log file.
  #
  exec {stderr}>&2
  exec 2>>"$log"

  function exit_trap ()
  {
    local status="$?"

    # Dump the log to stderr if exiting with non-zero status or verbose.
    #
    if [ $status -ne 0 -o ! "$quiet" ]; then

      # Keep the log if failed to dump for any reason.
      #
      if ! cat "$log" >&$stderr; then
        return
      fi
    fi

    rm -f "$log"
  }

  trap exit_trap EXIT
fi

# Source the configurations.
#
for c in "${configurations[@]}"; do
  source "$c" >&2
done

# Make sure the commit file is present.
#
published_commit="$repo_dir.publish"
touch "$published_commit"

# Open the reading file descriptor and lock the repository. Fail if unable to
# lock before timeout.
#
exec {cfd}<"$published_commit"

if ! flock -w "$lock_timeout" "$cfd"; then
  info "another instance is already running"
  exit 2
fi

# Pull the repository.
#
# Git doesn't support the connection timeout option. The options we use are
# just an approximation of the former, that, in particular, don't cover the
# connection establishing. To work around this problem, before running a git
# command that assumes the remote repository communication we manually check
# connectivity with the remote repository.
#
if ! remote_url="$(git -C "$repo_dir" config --get remote.origin.url)"; then
  error "'$repo_dir' is not a git repository"
fi

run check_git_connectivity "$remote_url" "$timeout"

# Fail if no network activity happens during the time specified.
#
run git -c http.lowSpeedLimit=1 -c "http.lowSpeedTime=$timeout" \
-C "$repo_dir" pull -v >&2

# Match the HEAD commit id to the one stored in the file. If it matches, then
# nothing changed in the repository from the previous run and we can silently
# bail out.
#
commit="$(git -C "$repo_dir" rev-parse HEAD)"
pc="$(cat <&"$cfd")"

if [ "$commit" == "$pc" ]; then
  quiet=true
  exit 0
fi

# If bpkg path is not specified, then use the bpkg program from the script
# directory, if present. Otherwise, use the 'bpkg' path.
#
if [ -z "$bpkg" ]; then
  bpkg="$(dirname "$(realpath "${BASH_SOURCE[0]}")")/bpkg"

  if [ ! -x "$bpkg" ]; then
    bpkg=bpkg
  fi
fi

# Find repository sections.
#
manifests="$(find "$repo_dir/$repo_ver" -type f -name repositories.manifest)"

# Update the repository sections.
#
while read f; do
  run "$bpkg" rep-create "${rep_create_options[@]}" "$(dirname "$f")"
done <<<"$manifests"

# rsync (over ssh) the repository to the destinations.
#
# Approximate the data transfer timeout via the ServerAlive* ssh options,
# rounding the timeout up to the nearest multiple of ten.
#
# Note: must not contain spaces/use quoting (see rsync -e option).
#
n=$(($timeout > 0 ? ($timeout + 9) / 10 : 1))
ssh_options=(-o ConnectTimeout=$timeout \
	     -o ServerAliveInterval=10 \
	     -o ServerAliveCountMax=$n)

for d in "${destinations[@]}"; do

  # -r (recursive)
  # -l (copy symlinks and symlinks)
  # -p (preserve perms)
  # -t (preserve timestamps)
  # -O (omit dir timestamps)
  #
  # -c (use checksum)
  # -e (remote shell command)
  #
  # --safe-links       (ignore symlinks pointing outside the tree)
  # --delay-updates    (first upload all files on the side then move)
  # --prune-empty-dirs (remove empty dirs)
  # --delete-after     (delete entries after the transfer)
  #
  # We also exclude hiddent files (start with dot).
  #
  run rsync -v -rlptO -c --safe-links --delay-updates --exclude '.*' \
--prune-empty-dirs --delete-after -e "ssh ${ssh_options[*]}" \
"${rsync_options[@]}" "$repo_dir/$repo_ver/" "$d/" >&2

done

echo "$commit" >"$published_commit"