/**
 * models/engine/wordcount.ts — the word count that enforces the plan limits.
 *
 * DOCUMENTED ALGORITHM (the same one runs in the browser, in public/app.js, for
 * the live counter; the server's count is the one that decides):
 *
 *  - UNICODE source text: Unicode word segmentation (UAX #29) through
 *    Intl.Segmenter with the Odia locale ("or"), counting only segments the
 *    segmenter marks as word-like. Odia vowel signs, virama and nukta stay inside
 *    their word, and punctuation such as the danda is not a word.
 *    Fallback when Intl.Segmenter is unavailable: split on Unicode whitespace and
 *    count pieces that contain at least one letter, mark or digit (\p{L}\p{M}\p{N}).
 *
 *  - LEGACY source text: legacy Odia fonts re-use ASCII and Latin-1 punctuation
 *    (@ [ ] \ ^ _ ` ~ { | } and others) as Odia glyphs, so Unicode segmentation
 *    would split one Odia word into several or drop parts of it. A word is
 *    therefore any run of characters between whitespace.
 *
 * Whitespace means what JavaScript's \s matches (spaces, tabs, line breaks,
 * no-break space and the other Unicode space separators).
 */

export type CountMode = "unicode" | "legacy";

let segmenter: Intl.Segmenter | null | undefined;
function getSegmenter(): Intl.Segmenter | null {
  if (segmenter === undefined) {
    try {
      segmenter = typeof Intl !== "undefined" && "Segmenter" in Intl ? new Intl.Segmenter("or", { granularity: "word" }) : null;
    } catch {
      segmenter = null;
    }
  }
  return segmenter;
}

export function countWords(text: string, mode: CountMode): number {
  if (!text) return 0;
  if (mode === "legacy") {
    const m = text.match(/\S+/gu);
    return m ? m.length : 0;
  }
  const seg = getSegmenter();
  if (seg) {
    let n = 0;
    for (const s of seg.segment(text)) if (s.isWordLike) n++;
    return n;
  }
  return text.split(/\s+/u).filter((w) => /[\p{L}\p{M}\p{N}]/u.test(w)).length;
}