import { stateFile } from "./paths";
import { readJson, writeJsonAtomic, withStateLock } from "./state";

// The word selector's vocabulary.
//
// This replaces the free-text "other" note, which was the only way to say what a
// clip actually contained and was useless downstream: one 284-clip pass produced
// "do uh", "to uh", "when uhh", "uhh literally", "duuhh", "uh-huh" and three
// separate clean "and"s -- all prose, none of it filterable by the arranger.
//
// CURATED comes first because it covers most clips in one keystroke. COMPOUNDS
// are the two-word forms that turned up repeatedly in those notes. Anything else
// gets typed once and joins vocab.json for every later clip.

export const CURATED = ["um", "uh", "umm", "uhh", "er", "erm", "mm-hmm", "uh-huh"];

export const COMPOUNDS = [
  "and um",
  "um and",
  "the um",
  "and uh",
  "uh and",
  "the uh",
  "to uh",
  "do uh",
  "a um",
  "in the um",
  "for the uh",
];

export type Vocab = { words: string[] };

export async function readVocab(): Promise<string[]> {
  const v = await readJson<Vocab>(stateFile("vocab.json"), { words: [] });
  return v.words ?? [];
}

/** Add a typed word to the persisted list. Idempotent, case-insensitively. */
export async function addVocab(word: string): Promise<string[]> {
  const w = word.trim();
  if (!w) return readVocab();
  return withStateLock(async () => {
    const v = await readJson<Vocab>(stateFile("vocab.json"), { words: [] });
    const words = v.words ?? [];
    const known = new Set([...CURATED, ...COMPOUNDS, ...words].map((s) => s.toLowerCase()));
    if (!known.has(w.toLowerCase())) {
      words.push(w);
      await writeJsonAtomic(stateFile("vocab.json"), { words });
    }
    return words;
  });
}