#!/usr/bin/env bash
#
# Refresh the vendored copy of Digital Extremes' drop tables.
#
# /Warframe reads Blog/Resources/droptables.html rather than warframe.com,
# which blocks requests from the server. This runs before a publish so each
# deploy ships current data. Commit the result.
#
# A failed refresh is a warning, not an error: the committed copy is still good
# data, and an outage at DE should not block a deploy that has nothing to do
# with Warframe. Missing the file entirely is an error, because then there is
# nothing to publish.

set -uo pipefail

url="${WARFRAME_DROPTABLES_URL:-https://www.warframe.com/droptables}"
repo="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
target="$repo/Blog/Resources/droptables.html"
tmp="$(mktemp)"
trap 'rm -f "$tmp"' EXIT

de_last_update() {
    grep -A1 '<b>Last Update:</b>' "$1" 2>/dev/null | tail -1 | tr -d '\r'
}

keep_existing() {
    echo "WARNING: could not refresh the drop tables: $1" >&2

    if [ ! -f "$target" ]; then
        echo "ERROR: and there is no committed copy at $target to fall back on." >&2
        exit 1
    fi

    echo "WARNING: publishing the committed copy (DE last update $(de_last_update "$target"))." >&2
    exit 0
}

echo "Fetching $url"
curl -sSL --fail --max-time 180 -o "$tmp" "$url" || keep_existing "download failed"

# The publish that follows ships whatever lands here, so an error page or a
# truncated download must not replace a copy that is known good. The real page
# is around 4 MB across twenty sections; anything far short of that is not it.
bytes=$(wc -c < "$tmp" | tr -d ' ')
sections=$(grep -c '<h3 id=' "$tmp")

if [ "$bytes" -lt 1000000 ] || [ "$sections" -lt 15 ]; then
    keep_existing "got $bytes bytes and $sections sections, which is not the drop table page"
fi

mv "$tmp" "$target"
trap - EXIT

echo "Updated Blog/Resources/droptables.html: $bytes bytes, $sections sections, DE last update $(de_last_update "$target")"