#!/usr/bin/env bash
# Build the site (web/scripts/build.sh) and push it to a ref-named path in
# infra's site bucket.
#
#   web/scripts/push.sh
#
# BUCKET overrides the bucket; the default is the name infra builds from
# <project>-web-bucket-<env>-<account id>, all of which are fixed.
#
# The push changes nothing the site serves. CloudFront serves whatever path
# infra's site_origin_path points at; to serve this build, set that variable
# to the path this script prints, apply, and invalidate the distribution
# (infra's site_distribution_id output).
set -euo pipefail

BUCKET="${BUCKET:-lance-blue-web-bucket-prod-179302349187}"

: "${AWS_PROFILE:?set AWS_PROFILE; default credentials are almost never the right account}"

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

BUILD_REF="$(scripts/build-ref.sh)"
export BUILD_REF
prefix="releases/$BUILD_REF"

web/scripts/build.sh

cd web

# Hashed assets never change under their names, so browsers may keep them
# forever. Everything else keeps its URL across deploys and must not be
# cached by browsers at all.
aws s3 sync dist/assets "s3://$BUCKET/$prefix/assets" --delete \
  --cache-control "public, max-age=31536000, immutable"
aws s3 sync dist "s3://$BUCKET/$prefix" --delete --exclude "assets/*" \
  --cache-control "no-cache"

# CloudFront looks for index.html at the root and nowhere else, so every
# prerendered page needs a second copy under the address the site links to:
# dist/blog/index.html is also the key "blog". Skipping this does not produce a
# missing page, it produces the wrong one - both 403 and 404 come back as the
# app's own index.html with a 200.
#
# The content type has to be given. These keys have no extension for the CLI to
# guess from, and what it guesses instead is a download.
while IFS= read -r page; do
  key="${page#dist/}"
  aws s3 cp "$page" "s3://$BUCKET/$prefix/${key%/index.html}" \
    --content-type "text/html; charset=utf-8" \
    --cache-control "no-cache"
done < <(find dist -mindepth 2 -name index.html)

# The feed does have an extension, and what the CLI guesses from it is
# application/xml. Readers take that, but it is not what the <link> on every
# page says the document is, and the two disagreeing is the sort of thing that
# is only ever noticed by whichever reader is strict about it.
if [ -f dist/atom.xml ]; then
  aws s3 cp dist/atom.xml "s3://$BUCKET/$prefix/atom.xml" \
    --content-type "application/atom+xml; charset=utf-8" \
    --cache-control "no-cache"
fi

# The same fix for the standard.site key, which has no extension either and
# has to arrive as text rather than as a download. Absent until `sequoia init`
# writes it, so this is a no-op until the publication record exists.
wellknown="dist/.well-known/site.standard.publication"
if [ -f "$wellknown" ]; then
  aws s3 cp "$wellknown" "s3://$BUCKET/$prefix/${wellknown#dist/}" \
    --content-type "text/plain; charset=utf-8" \
    --cache-control "no-cache"
fi

echo
echo "Pushed to s3://$BUCKET/$prefix"
echo "To serve it: set site_origin_path = \"/$prefix\" in infra, apply,"
echo "then invalidate CloudFront."