import { readFile } from 'node:fs/promises';
import pg from 'pg';
import { storedGroup, tfdaGroup } from '../groups.ts';
import { parseCsv } from './cnf.ts';
import { importDrafts, rowObject, toNumber } from './national.ts';
import type { Cell, Draft, NutrientKey, Parsed } from './national.ts';
import { openZip } from './xlsx.ts';

// Taiwan Food Composition Database ("Nutrition Information Database", Food and Drug Administration, TFDA),
// SRC-20. Open data under the Open Government Data License, version 1.0 (data.gov.tw/en/datasets/8543),
// which grants a "perpetual, worldwide, non-exclusive, irrevocable, royalty-free copyright license ... for any
// purpose, including ... Derivative Works either as products or services" and requires the attribution of
// its Exhibit. The CSV (zipped, one long table: one row per food and analysis item) is the official export.
export const tfdaSourceName = 'Nutrition Information Database, Food and Drug Administration, Taiwan (TFDA)';
export const tfdaAttribution = 'Food and Drug Administration, Ministry of Health and Welfare, Taiwan, 2026, Nutrition Information Database. '
  + 'The Open Data is made available to the public under the Open Government Data License, version 1.0: https://data.gov.tw/license';

const columns = ['食品分類', '資料類別', '整合編號', '樣品名稱', '俗名', '樣品英文名稱', '內容物描述', '廢棄率', '分析項分類',
  '分析項', '含量單位', '每100克含量', '樣本數', '標準差', '每單位含量', '每單位重', '每單位重含量'];

// Analysis item + unit -> key. "總碳水化合物" is carbohydrate by difference, so it includes fibre (FD-11 `total`).
const items: Record<string, [NutrientKey, string]> = {
  '熱量': ['enerc_kcal', 'kcal'], '粗蛋白': ['procnt_g', 'g'], '粗脂肪': ['fat_g', 'g'], '總碳水化合物': ['chocdf_g', 'g'],
  '膳食纖維': ['fibtg_g', 'g'], '糖質總量': ['sugar_g', 'g'], '飽和脂肪': ['fasat_g', 'g'], '鈉': ['na_mg', 'mg'],
  '膽固醇': ['chole_mg', 'mg'], '鉀': ['k_mg', 'mg'], '鈣': ['ca_mg', 'mg'], '鐵': ['fe_mg', 'mg'],
  '維生素D總量(ug)': ['vitd_ug', 'ug'], '酒精含量': ['alc_g', 'g'], '水分': ['water_g', 'g'],
};
// The table gives a sample count (樣本數) for every value it analysed and none for the values it computes from
// them; of those only these three are read, and they are `calculated`. Any other value without a sample count
// stays unknown (FD-07).
const computedItems = new Set(['熱量', '總碳水化合物', '飽和脂肪', '糖質總量']);

const clean = (cell: Cell | undefined) => (cell ?? '').replace(/\s+/g, ' ').trim();

export function parseTfda(zipBytes: Buffer): Parsed {
  const archive = openZip(zipBytes);
  const file = archive.names.find(name => name.toLowerCase().endsWith('.csv'));
  const bytes = file ? archive.read(file) : null;
  if (!bytes) throw new Error('TFDA zip has no CSV');
  const [header, ...rows] = parseCsv(bytes);
  if (!header || columns.some((title, at) => header[at]?.trim() !== title)) throw new Error('TFDA columns changed');
  const byFood = new Map<string, Cell[][]>();
  for (const row of rows) {
    const id = clean(row[2]);
    if (id) byFood.set(id, [...(byFood.get(id) ?? []), row]);
  }
  const out: Parsed = { drafts: [], skipped: [] };
  for (const [id, list] of byFood) {
    const head = list[0];
    // The English field lists synonyms joined by ; or 、; the first is the name, the rest are kept.
    const english = clean(head[5]).split(/\s*[;；、]\s*/).filter(Boolean);
    const native = clean(head[3]);
    const name = english[0] ?? native;
    if (!name) { out.skipped.push({ recordId: id, reason: 'no name' }); continue; }
    const values: Draft['values'] = {};
    const kinds: Draft['kinds'] = {};
    const seen = new Set<string>();
    let broken = false;
    for (const row of list) {
      const known = items[clean(row[9])];
      if (!known) continue;
      if (clean(row[10]) !== known[1]) throw new Error(`TFDA unit of ${clean(row[9])} changed`);
      if (seen.has(known[0])) { broken = true; break; }
      seen.add(known[0]);
      const value = toNumber(clean(row[11]));
      const count = toNumber(clean(row[12]));
      const type = count !== null && count >= 1 ? 'analytic' : computedItems.has(clean(row[9])) ? 'calculated' : null;
      if (value === null || value < 0 || !type) continue;
      values[known[0]] = value; kinds[known[0]] = type;
    }
    if (broken) { out.skipped.push({ recordId: id, reason: 'duplicate analysis item' }); continue; }
    if (!Object.keys(values).length) { out.skipped.push({ recordId: id, reason: 'no composition rows' }); continue; }
    out.drafts.push({
      recordId: id, name, basis: '100g', carbohydrateDefinition: 'total', values, kinds, measures: [],
      group: storedGroup(tfdaGroup(clean(head[0]))),
      added: { name_language: english.length ? 'en' : 'zh', native_name: native || null, native_language: 'zh',
        synonyms_en: english.slice(1), tfda_category: clean(head[0]), tfda_record_class: clean(head[1]) },
      raw: Buffer.from(JSON.stringify(list.map(row => rowObject(header, row)))),
    });
  }
  return out;
}

export async function importTfdaFile(client: pg.Client, path: string, releasedAt: string) {
  return importDrafts(client, { src: 'SRC-20', market: 'TW', family: 'open' },
    parseTfda(await readFile(path)), releasedAt, tfdaSourceName);
}