#!/usr/bin/env bash
# Publish the updates to ATProto as standard.site records: one
# site.standard.document per post, in the publication owner's own repository.
# Configuration is web/sequoia.json.
#
#   ATP_IDENTIFIER=... ATP_APP_PASSWORD=... web/scripts/publish.sh [--dry-run]
#
# The last step of a deploy, not part of one. A document record says "this
# post is at this address", so the address has to answer first — otherwise
# the network is told about a page that is not there. This refuses to publish
# a post the live site is not already serving.
#
# ATP_APP_PASSWORD is an app password, never an account password; it is
# never written or printed here.
set -euo pipefail

cd "$(dirname "$0")/.."

dry_run=""
if [ $# -gt 0 ]; then
  case "$1" in
    --dry-run | -n) dry_run=1 ;;
    *)
      echo "usage: web/scripts/publish.sh [--dry-run]" >&2
      exit 2
      ;;
  esac
fi

sequoia="./node_modules/.bin/sequoia"
[ -x "$sequoia" ] || {
  echo "$sequoia is missing. Run: npm --prefix web install" >&2
  exit 1
}

if grep -q PLACEHOLDER sequoia.json; then
  echo "sequoia.json still has the placeholder publicationUri." >&2
  echo "Create the publication once (sequoia login && sequoia init)," >&2
  echo "then put its at:// URI there and in" >&2
  echo "public/.well-known/site.standard.publication." >&2
  exit 1
fi

: "${ATP_APP_PASSWORD:?set it to an app password for the publishing account}"
export ATP_APP_PASSWORD
export ATP_IDENTIFIER="${ATP_IDENTIFIER:?set it to the publishing account handle}"

# What publishing would do, and the only place the post URLs come from. A
# real run may skip some of these after syncing state from the PDS, so this
# is a superset of what gets written — the safe direction for a check.
plan="$("$sequoia" publish --dry-run --verbose)"
echo "$plan"

# Read from the config rather than spelled again here: a second copy is a
# second thing to get wrong.
prefix="$(node -p 'const c = require("./sequoia.json"); c.siteUrl + c.pathPrefix')"
urls="$(printf '%s\n' "$plan" | grep -oE "${prefix}/[^[:space:]]+" | sort -u || true)"
if [ -z "$urls" ]; then
  echo "Nothing to publish."
  exit 0
fi

# The canonical link is what proves the right page is being served at the
# claimed address, not the status code.
echo
echo "Checking the live site..."
not_live=0
while IFS= read -r url; do
  if ! body="$(curl -fsS --max-time 20 "$url")"; then
    echo "  not serving: $url" >&2
    not_live=1
  elif ! printf '%s' "$body" | grep -qF "rel=\"canonical\" href=\"$url\""; then
    echo "  wrong page at: $url" >&2
    not_live=1
  else
    echo "  live: $url"
  fi
done <<<"$urls"

if [ "$not_live" -ne 0 ]; then
  echo >&2
  echo "Refusing to publish: deploy first, then run this again." >&2
  exit 1
fi

if [ -n "$dry_run" ]; then
  echo
  echo "Dry run: every post is live, nothing published."
  exit 0
fi

echo
"$sequoia" publish

echo
echo "Published. Commit the atUri frontmatter sequoia wrote into the posts."