import { test } from 'node:test';
import assert from 'node:assert/strict';
import type { Word } from './whisper.ts';
import { wordsFromJson } from './whisper.ts';
import { assColor, assTime, buildAss, buildSrt, isNoise, toLines } from './subtitles.ts';

function w(text: string, start: number, end: number): Word {
  return { text, start, end };
}

// ------------------------------------------------------------ Wortzerlegung

test('setzt Wortstuecke wieder zusammen', () => {
  // So sieht whisper.cpp wirklich aus: Tokens mit fuehrendem Leerzeichen
  // beginnen ein Wort, alles andere haengt sich an.
  const json = JSON.stringify({
    transcription: [{
      offsets: { from: 0, to: 4200 },
      tokens: [
        { text: '[_BEG_]', offsets: { from: 0, to: 0 } },
        { text: ' Sav', offsets: { from: 2280, to: 3120 } },
        { text: 'ina', offsets: { from: 3280, to: 3880 } },
        { text: ',', offsets: { from: 3990, to: 4140 } },
      ],
    }],
  });
  const words = wordsFromJson(json);
  assert.deepEqual(words.map((x) => x.text), ['Savina,']);
  assert.equal(words[0]?.start, 2.28);
});

test('wirft Sondertokens weg', () => {
  const json = JSON.stringify({
    transcription: [{ tokens: [
      { text: '[_BEG_]', offsets: { from: 0, to: 0 } },
      { text: ' hallo', offsets: { from: 100, to: 400 } },
      { text: '[_TT_120]', offsets: { from: 400, to: 400 } },
    ] }],
  });
  assert.deepEqual(wordsFromJson(json).map((x) => x.text), ['hallo']);
});

test('zieht rueckwaerts laufende Zeitstempel gerade', () => {
  const json = JSON.stringify({
    transcription: [{ tokens: [
      { text: ' eins', offsets: { from: 1000, to: 2000 } },
      { text: ' zwei', offsets: { from: 1500, to: 1800 } },
    ] }],
  });
  const words = wordsFromJson(json);
  assert.ok(words[1]!.start >= words[0]!.end, 'zweites Wort darf nicht vor dem ersten enden');
  assert.ok(words[1]!.end > words[1]!.start);
});

// ------------------------------------------------------------ Zeilenbildung

test('bricht lange Rede in kurze Zeilen', () => {
  const words = ['eins', 'zwei', 'drei', 'vier', 'fuenf', 'sechs', 'sieben', 'acht']
    .map((t, i) => w(t, i * 0.4, i * 0.4 + 0.35));
  const lines = toLines(words);
  assert.ok(lines.length >= 2, 'acht Woerter gehoeren nicht auf eine Zeile');
  for (const l of lines) assert.ok(l.words.length <= 5, 'Zeile zu lang');
});

test('haengt einzeln stehende Woerter an den Nachbarn', () => {
  // Genau der Fall aus dem echten Clip: langsames Sprechen erzeugte
  // Zeilen mit einem Wort ("Dank", "es,", "Savina,")
  const words = [w('Dank', 0, 1.14), w('es,', 1.14, 2.28), w('Savina,', 2.28, 4.14)];
  const lines = toLines(words);
  assert.ok(lines.length < 3, `immer noch zersplittert: ${JSON.stringify(lines.map(l => l.words.map(x => x.text)))}`);
  for (const l of lines) {
    assert.ok(l.words.length > 1 || lines.length === 1, 'Zeile mit nur einem Wort uebrig');
  }
});

test('trennt bei laengeren Sprechpausen', () => {
  const words = [w('vor', 0, 0.4), w('der', 0.4, 0.8), w('Pause', 4.0, 4.4)];
  const lines = toLines(words);
  assert.equal(lines.length, 2);
  assert.equal(lines[1]?.words[0]?.text, 'Pause');
});

test('filtert Geraeuschbeschreibungen', () => {
  assert.equal(isNoise('*Lachen*'), true);
  assert.equal(isNoise('[Musik]'), true);
  assert.equal(isNoise('(Applaus)'), true);
  assert.equal(isNoise('Lachen'), false);
  const lines = toLines([w('hallo', 0, 0.5), w('*Lachen*', 0.6, 1.2)]);
  assert.deepEqual(lines.flatMap((l) => l.words.map((x) => x.text)), ['hallo']);
});

// ------------------------------------------------------------ ASS-Ausgabe

test('Zeilen ueberlappen sich nie', () => {
  // Zwei gleichzeitig sichtbare Dialoge stapelt ASS uebereinander
  const words = Array.from({ length: 12 }, (_, i) => w(`w${i}`, i * 0.5, i * 0.5 + 0.45));
  const ass = buildAss(toLines(words), { width: 1080, height: 1920 });
  const times = [...ass.matchAll(/^Dialogue: 0,([\d:.]+),([\d:.]+),/gm)]
    .map((m) => [toSeconds(m[1]!), toSeconds(m[2]!)] as const);
  assert.ok(times.length >= 2);
  for (let i = 1; i < times.length; i++) {
    assert.ok(times[i]![0] >= times[i - 1]![1] - 0.011,
      `Zeile ${i} beginnt bei ${times[i]![0]}, vorherige endet erst bei ${times[i - 1]![1]}`);
  }
});

test('jede Zeile bleibt mindestens kurz stehen', () => {
  const ass = buildAss(toLines([w('kurz', 0, 0.05)]), { width: 1080, height: 1920 });
  const m = /^Dialogue: 0,([\d:.]+),([\d:.]+),/m.exec(ass);
  assert.ok(m);
  assert.ok(toSeconds(m[2]!) - toSeconds(m[1]!) >= 0.29);
});

test('Kopf enthaelt Aufloesung, Stil und Sicherheitsabstand', () => {
  const ass = buildAss(toLines([w('test', 0, 1)]), { width: 1080, height: 1920 });
  assert.match(ass, /PlayResX: 1080/);
  assert.match(ass, /PlayResY: 1920/);
  // Alignment 2 = unten mittig, MarginV 340 haelt die Bedienoberflaeche frei
  assert.match(ass, /,2,70,70,340,1$/m);
});

test('Farben werden in die ASS-Reihenfolge gedreht', () => {
  // RRGGBB -> &HAABBGGRR
  assert.equal(assColor('FFD400'), '&H0000D4FF');
  assert.equal(assColor('FFFFFF'), '&H00FFFFFF');
  assert.equal(assColor('000000', 0x80), '&H80000000');
});

test('Zeitformat entspricht ASS', () => {
  assert.equal(assTime(0), '0:00:00.00');
  assert.equal(assTime(61.5), '0:01:01.50');
  assert.equal(assTime(3661.239), '1:01:01.24');
  // Aufrundung auf 100 Hundertstel darf nicht ".100" ergeben
  assert.equal(assTime(59.999), '0:01:00.00');
});

test('jedes Wort bekommt eine Karaoke-Marke', () => {
  const ass = buildAss(toLines([w('eins', 0, 0.5), w('zwei', 0.5, 1.0)]),
    { width: 1080, height: 1920 });
  const line = /^Dialogue:.*,,(.*)$/m.exec(ass)?.[1] ?? '';
  assert.equal((line.match(/\{\\k\d+\}/g) ?? []).length, 2);
});

test('SRT enthaelt denselben Text', () => {
  const lines = toLines([w('hallo', 0, 0.5), w('welt', 0.5, 1.0)]);
  const srt = buildSrt(lines);
  assert.match(srt, /hallo welt/);
  assert.match(srt, /00:00:00,000 --> /);
});

function toSeconds(t: string): number {
  const [h, m, s] = t.split(':');
  return Number(h) * 3600 + Number(m) * 60 + Number(s);
}