#!/usr/bin/env bash
# Updates the server (granstrm, from the srv repo) the way nixos-update updates this
# desktop, from a throwaway worktree so the checkout you edit in is left alone.
# Started from the "Server" half of the control centre's update tile in a kitty
# window (installed as `server-update`, modules/home/packages.nix).
#
#   1. a temporary worktree of main
#   2. nix flake update, committed there
#   3. builds the server's system and the repo's VM tests (router, web, media), then
#      nvd's list of what changed against the system the server runs
#   4. asks before going on; then pushes the commit, switches the server to it over
#      SSH (nixos-rebuild switch --target-host) and brings your checkout's main up
#      to date
#   5. if the kernel changed, asks whether to reboot the server now. It is the
#      house router, so the internet drops while it reboots.
#
# If anything fails, the window stays open until you press Enter, and nothing on the
# server is changed.
set -E

repo=${SRV_REPO:-$HOME/git/srv}
host=${SRV_HOST:-granstrm}
tmp=""

step() { printf '\n\e[1;34m==> %s\e[0m\n' "$*"; }
note() { printf '\e[1;33m%s\e[0m\n' "$*"; }
wait_close() { read -rp "Press Enter to close. " _ || true; }

die() {
  trap - ERR
  printf '\n\e[1;31m%s\e[0m\n' "$*"
  wait_close
  exit 1
}

cleanup() {
  if [ -n "$tmp" ]; then
    git -C "$repo" worktree remove --force "$tmp" 2>/dev/null || true
    rm -rf "$tmp"
    tmp=""
  fi
}

trap 'die "The update failed (see above). The server was not changed."' ERR
trap cleanup EXIT

step "Fetching $repo"
git -C "$repo" worktree prune
git -C "$repo" fetch origin main
git -C "$repo" merge-base --is-ancestor origin/main main ||
  die "main in $repo is behind origin/main (or has diverged); pull first."
base=$(git -C "$repo" rev-parse main)
ahead=$(git -C "$repo" rev-list --count origin/main..main)
dirty=$(git -C "$repo" status --porcelain)

tmp=$(mktemp -d -t srv-update.XXXXXX)
step "Creating a worktree of main in $tmp"
git -C "$repo" worktree add --detach "$tmp" main
cd "$tmp" || die "Could not enter $tmp."

step "Updating flake inputs"
nix flake update --commit-lock-file
if [ "$(git rev-parse HEAD)" = "$base" ]; then
  printf '\nEverything is already up to date.\n'
  wait_close
  exit 0
fi

step "Building the server's system"
nixos-rebuild build --flake "$tmp#$host"

checks=$(nix eval --raw .#checks.x86_64-linux --apply 'c: builtins.concatStringsSep " " (builtins.attrNames c)')
step "Running the VM tests ($checks)"
for c in $checks; do
  nix build --no-link ".#checks.x86_64-linux.$c"
done

step "What changes"
git --no-pager log -1 --format=%b
running=$(ssh "$host" readlink -f /run/current-system)
# Usually still here from the last deploy; fetched from the server if not.
nix path-info "$running" >/dev/null 2>&1 || nix copy --from "ssh-ng://$host" "$running"
nvd diff "$running" "$tmp/result"

echo
[ -z "$dirty" ] || note "Uncommitted changes in $repo are not part of this update."
[ "$ahead" -eq 0 ] || note "main has $ahead commit(s) not on GitHub yet; they are pushed and deployed too."
read -rp "Commit, push and switch the server? [y/N] " answer || answer=""
case $answer in
  [yY]*) ;;
  *)
    echo "Nothing was changed."
    sleep 1
    exit 0
    ;;
esac

step "Pushing"
git push origin HEAD:main

step "Switching the server"
nixos-rebuild switch --flake "$tmp#$host" --target-host "$host"

step "Updating main in $repo"
commit=$(git rev-parse HEAD)
if [ "$(git -C "$repo" symbolic-ref --quiet --short HEAD)" = main ]; then
  git -C "$repo" merge --ff-only "$commit"
else
  git -C "$repo" update-ref refs/heads/main "$commit" "$base"
fi
cd "$repo"
cleanup

# The same test NixOS's own auto-upgrade uses: a new kernel or initrd only takes
# effect after a reboot.
# shellcheck disable=SC2016 # expanded on the server
if ssh "$host" '[ "$(readlink /run/booted-system/{initrd,kernel,kernel-modules})" = "$(readlink /run/current-system/{initrd,kernel,kernel-modules})" ]'; then
  printf '\nDone. No reboot needed.\n'
  wait_close
  exit 0
fi

echo
note "The kernel changed; it is used after the server's next reboot."
read -rp "Reboot the server now? The internet drops for a few minutes. [y/N] " answer || answer=""
case $answer in
  [yY]*)
    step "Rebooting the server"
    ssh "$host" systemctl reboot || true
    ;;
  *) echo "Not rebooted. The new kernel is used after the next reboot." ;;
esac
wait_close