// The chunker: paragraph packing, hard splits of over-long paragraphs, positions a citation can
// point at. Pure paths — no store, no model.

import { describe, expect, it } from "vitest"
import { CHUNK_SIZE, chunkText } from "../core/chunk.js"

describe("chunkText", () => {
  it("keeps a short document as one chunk", () => {
    const chunks = chunkText("One paragraph.\n\nTwo paragraphs.")
    expect(chunks).toHaveLength(1)
    expect(chunks[0]?.position).toBe(1)
    expect(chunks[0]?.text).toContain("Two paragraphs.")
  })

  it("splits on paragraph boundaries when the pack overflows the size", () => {
    const paragraph = "word ".repeat(500) // 2500 characters
    const chunks = chunkText(`${paragraph.trim()}\n\n${paragraph.trim()}`)
    expect(chunks.length).toBeGreaterThanOrEqual(2)
    for (const chunk of chunks) expect(chunk.text.length).toBeLessThanOrEqual(CHUNK_SIZE)
    expect(chunks.map((chunk) => chunk.position)).toEqual(chunks.map((_, at) => at + 1))
  })

  it("hard-splits a paragraph longer than the size", () => {
    const long = "x".repeat(CHUNK_SIZE * 2 + 100)
    const chunks = chunkText(long)
    expect(chunks.length).toBeGreaterThan(2)
    expect(chunks.every((chunk) => chunk.text.length <= CHUNK_SIZE)).toBe(true)
  })

  it("refuses an empty document with a named field", () => {
    expect(() => chunkText("")).toThrow(/nothing to index/)
  })
})