#!/usr/bin/env bash
# Daily update of Taiwan stock CSVs from the exchanges' official open data (2 requests in total).
#
#   scripts/fetch_tw_daily.sh [-d outdir]              merge the latest day into each CODE.TW.csv
#   scripts/fetch_tw_daily.sh --snapshot DIR          write DIR/YYYY-MM-DD.csv (all stocks, one day)
#
# Downloads the latest trading day of every TWSE-listed and TPEx (OTC) security (2 requests).
# Merge mode (default outdir: data/tw) adds the rows to each existing CODE.TW.csv / CODE.TWO.csv
# (create those first with fetch.sh). Snapshot mode needs no per-stock files: it writes one
# "Symbol,Date,Open,High,Low,Close,Volume" file per trading day, which the VPS publishes for
# scripts/sync_from_vps.sh to apply. Only the latest day is available from these endpoints.
# Prices are raw, volume is in shares.
set -u
source "$(dirname "$0")/lib.sh"
outdir=data/tw; snapshot=
args=(); for a in "$@"; do [ "$a" = "--snapshot" ] && a=-s; args+=("$a"); done   # getopts has no long options
set -- ${args[@]+"${args[@]}"}
while getopts "d:s:" o; do
  case $o in d) outdir=$OPTARG;; s) snapshot=$OPTARG;; *) sed -n '2,12p' "$0"; exit 2;; esac
done
for cmd in curl jq; do command -v $cmd >/dev/null || { echo "$cmd is required" >&2; exit 1; }; done
[ -n "$snapshot" ] || [ -d "$outdir" ] || { echo "$outdir does not exist; run fetch.sh first" >&2; exit 1; }

# Private scratch folder (always a fresh mktemp directory); removed again when the script ends.
work=$(mktemp -d) && [ -n "$work" ] && [ -d "$work" ] || { echo "cannot create a temporary directory" >&2; exit 1; }
trap 'rm -rf "$work"' EXIT

get_json() {  # URL OUTFILE; retries until the response parses as a JSON array
  local try
  # The exchanges sometimes cut a large response short (curl exit 56) or return an empty array
  # before the day is published; wait longer between tries (2, 4, ... 30 s).
  for try in 1 2 3 4 5 6 7 8; do
    rc=0; curl -s --http1.1 -m 120 -A "Mozilla/5.0" "$1" -o "$2" || rc=$?
    if [ $rc -eq 0 ] && jq -e 'type == "array" and length > 0' "$2" >/dev/null 2>&1; then return 0; fi
    echo "  $1: bad response (try $try, curl exit $rc), retrying" >&2; sleep $(( (try * 2 > 30) ? 30 : try * 2 ))
  done
  echo "failed to download $1" >&2; return 1
}

# The jq programs below turn ROC dates (1151008 -> 2026-10-08) into ISO dates and keep only rows with
# positive numeric OHLC and volume (a missing value becomes null, never a dropped field), emitting "SYMBOL,Date,O,H,L,C,V".
rows="$work/rows.csv"; : > "$rows"
get_json "https://openapi.twse.com.tw/v1/exchangeReport/STOCK_DAY_ALL" "$work/twse.json" || exit 1
get_json "https://www.tpex.org.tw/openapi/v1/tpex_mainboard_daily_close_quotes" "$work/tpex.json" || exit 1

jq -r '
  def num: if type == "string" then (gsub(",";"") | if test("^[0-9]+(\\.[0-9]+)?$") then tonumber else null end) else null end | if . != null and . > 0 then . else null end;
  .[] | ((.Date[0:3] | tonumber + 1911 | tostring) + "-" + .Date[3:5] + "-" + .Date[5:7]) as $d
  | [(.Code + ".TW"), $d, (.OpeningPrice|num), (.HighestPrice|num), (.LowestPrice|num), (.ClosingPrice|num), (.TradeVolume|num)]
  | select(all(.[2:7][]; . != null)) | @csv' "$work/twse.json" | tr -d '"' >> "$rows"
jq -r '
  def num: if type == "string" then (gsub(",";"") | if test("^[0-9]+(\\.[0-9]+)?$") then tonumber else null end) else null end | if . != null and . > 0 then . else null end;
  .[] | ((.Date[0:3] | tonumber + 1911 | tostring) + "-" + .Date[3:5] + "-" + .Date[5:7]) as $d
  | [(.SecuritiesCompanyCode + ".TWO"), $d, (.Open|num), (.High|num), (.Low|num), (.Close|num), (.TradingShares|num)]
  | select(all(.[2:7][]; . != null)) | @csv' "$work/tpex.json" | tr -d '"' >> "$rows"

echo "TWSE date: $(jq -r '.[0].Date' "$work/twse.json")  TPEx date: $(jq -r '.[0].Date' "$work/tpex.json") (ROC calendar)"
echo "usable rows: $(wc -l < "$rows" | tr -d ' ')"
[ -s "$rows" ] || { echo "no usable rows in either response" >&2; exit 1; }

if [ -n "$snapshot" ]; then
  mkdir -p "$snapshot"
  # TWSE and TPEx publish at different times, so one date can arrive in two runs: merge by symbol
  # into an existing snapshot instead of overwriting it.
  for d in $(cut -d, -f2 "$rows" | sort -u); do
    f="$snapshot/$d.csv"
    if [ -s "$f" ]; then state="updated"; else state="new"; fi
    awk -F, -v d="$d" '
      FNR == 1 && FILENAME != ARGV[ARGC-1] { next }              # header of the existing snapshot
      FILENAME == ARGV[ARGC-1] { if ($2 == d) r[$1] = $0; next }  # new rows win
      { if (!($1 in r)) r[$1] = $0 }
      END { for (k in r) print r[k] }' "$f" "$rows" 2>/dev/null | sort > "$work/snap.body"
    if [ ! -s "$f" ]; then awk -F, -v d="$d" '$2 == d' "$rows" | sort > "$work/snap.body"; fi
    { echo "Symbol,Date,Open,High,Low,Close,Volume"; cat "$work/snap.body"; } > "$f.tmp"
    if [ -s "$f" ] && cmp -s "$f" "$f.tmp"; then state="unchanged"; rm -f "$f.tmp"; else mv "$f.tmp" "$f"; fi
    echo "$state: $f ($(wc -l < "$work/snap.body" | tr -d ' ') rows)"
  done
  exit 0
fi

updated=0; skipped=0
while IFS= read -r sym; do
  file="$outdir/$sym.csv"
  [ -s "$file" ] || { skipped=$((skipped + 1)); continue; }
  new="$work/new.csv"; grep "^$sym," "$rows" | cut -d, -f2- > "$new"
  [ -s "$new" ] || continue
  merge_into "$file" "$new" && updated=$((updated + 1))
done < <(cut -d, -f1 "$rows" | sort -u)
echo "updated $updated files ($skipped securities without a CSV in $outdir ignored)"