/**
 * Sentence utilities: reading pre-processed example sentence files and building
 * the character-level indexes used when generating subject data.
 *
 * Sentence files live at data/lang/{userLang}/sentences/{targetLang}.tsv.
 */

import { distinct } from '@std/collections/distinct'
import { parse } from '@std/csv/parse'
import type { Locale } from '$/enums.ts'
import { DATA_ROOT } from './fs.ts'

export interface Sentences {
  /** Maps sentence text → user-language translation. */
  bySentence: Record<string, string>
  /** Maps each individual character → all sentence texts that contain it. */
  byChar: Map<string, string[]>
  /** All sentence texts, in curriculum-sorted order (simplest first). */
  sorted: string[]
}

/**
 * Reads the pre-processed sentence TSV for a given user + target language pair.
 * Returns `{ bySentence, keys }`. Returns empty values if the file doesn't exist.
 */
export function readSentences(
  userLang: string,
  locale: Locale,
): { bySentence: Record<string, string>; keys: string[] } {
  let text = ''
  try {
    text = Deno.readTextFileSync(`${DATA_ROOT}lang/${userLang}/sentences/${locale}.tsv`)
  } catch {
    return { bySentence: {}, keys: [] }
  }
  const rows = parse(text, { separator: '\t', lazyQuotes: true })
  const bySentence: Record<string, string> = {}
  const keys = distinct(
    rows.map(([_id, value, _enId, translation]) => {
      bySentence[value] = translation
      return value
    }),
  )
  return { bySentence, keys }
}

export interface SentenceEntry {
  value: string
  translation: string
  level: number
}

/**
 * Reads the pre-processed sentence TSV (with optional level column) for a given
 * user + target language pair, returning only sentences that have a level assigned.
 * Used to generate the sentence game data files.
 */
export function readSentencesForGame(userLang: string, locale: string): SentenceEntry[] {
  let text = ''
  try {
    text = Deno.readTextFileSync(`${DATA_ROOT}lang/${userLang}/sentences/${locale}.tsv`)
  } catch {
    return []
  }
  const rows = parse(text, { separator: '\t', lazyQuotes: true })
  const result: SentenceEntry[] = []
  for (const [_id, value, _enId, translation, levelStr] of rows) {
    const level = parseInt(levelStr ?? '')
    if (value && translation && !isNaN(level)) {
      result.push({ value, translation, level })
    }
  }
  return result
}

/**
 * Loads and indexes sentences for a user + target language pair.
 * Builds `byChar` for fast per-character lookup when generating subject examples.
 */
export function loadSentences(userLang: string, targetLang: string): Sentences {
  const raw = readSentences(userLang, targetLang as Locale)
  const byChar = new Map<string, string[]>()
  for (const key of raw.keys) {
    for (const char of key) {
      if (!byChar.has(char)) byChar.set(char, [])
      byChar.get(char)!.push(key)
    }
  }
  return { bySentence: raw.bySentence, byChar, sorted: raw.keys }
}