// fixtures/wordcount.test.mjs — counting, under plain Node.
//
// countWords is the only part of the word count with any judgement in
// it, and it is the part a person will quote to somebody else, so it is
// the part worth pinning. The dialog around it is markup.

import { test } from 'node:test'
import assert from 'node:assert/strict'

import { Schema } from 'prosemirror-model'
import { EditorState } from 'prosemirror-state'
import { schema as basic } from 'prosemirror-schema-basic'
import { addListNodes } from 'prosemirror-schema-list'

import { countWords, counts } from './wordcount.mjs'

const schema = new Schema({
  nodes: addListNodes(basic.spec.nodes, 'paragraph block*', 'block'),
  marks: basic.spec.marks,
})

function docOf(...paragraphs) {
  return schema.node(
    'doc',
    null,
    paragraphs.map((text) =>
      text ? schema.node('paragraph', null, [schema.text(text)]) : schema.node('paragraph'),
    ),
  )
}

test('nothing is nothing', () => {
  assert.equal(countWords(''), 0)
  assert.equal(countWords('   '), 0)
  assert.equal(countWords(null), 0)
})

test('ordinary English', () => {
  assert.equal(countWords('one'), 1)
  assert.equal(countWords('one two three'), 3)
  assert.equal(countWords('  leading and trailing  '), 3)
  assert.equal(countWords('line\nbreaks\tand\ttabs'), 4)
})

// Punctuation is where whitespace-splitting and real segmentation
// diverge, and where a person notices: an em-dashed clause is two
// words, not one.
test('punctuation is not a word', () => {
  assert.equal(countWords('yes — no'), 2)
  assert.equal(countWords('one, two, three.'), 3)
  assert.equal(countWords('"quoted"'), 1)
})

// The reason this uses Intl.Segmenter at all. Whitespace-splitting
// finds no boundary in these at all and would report 1.
test('a language without spaces between words still counts', (t) => {
  if (typeof Intl === 'undefined' || typeof Intl.Segmenter !== 'function') {
    t.skip('this platform has no Intl.Segmenter, so only the whitespace fallback exists')
    return
  }
  assert.ok(countWords('今日は良い天気です') > 1, 'Japanese counted as one word')
  assert.ok(countWords('这是一个句子') > 1, 'Chinese counted as one word')
})

test('the document count walks every block', () => {
  const state = EditorState.create({ doc: docOf('one two', 'three four five'), schema })
  const c = counts(state)
  assert.equal(c.words, 5)
  assert.equal(c.paragraphs, 2)
})

// The blocks are joined with a space, so the last word of one paragraph
// and the first of the next are two words rather than one. Without
// that, "one" and "two" below would be counted as "onetwo".
test('a paragraph boundary is a word boundary', () => {
  const state = EditorState.create({ doc: docOf('one', 'two'), schema })
  assert.equal(counts(state).words, 2)
})

test('characters are counted with and without spaces', () => {
  const state = EditorState.create({ doc: docOf('ab cd'), schema })
  const c = counts(state)
  assert.equal(c.characters, 5)
  assert.equal(c.charactersNoSpaces, 4)
})

test('no selection means no selection count, not a zero', () => {
  // A zero would be indistinguishable from "nothing is selected", which
  // is why the dialog hides the row rather than showing 0.
  const state = EditorState.create({ doc: docOf('one two'), schema })
  assert.equal(counts(state).selection, null)
})