// SPDX-License-Identifier: LicenseRef-ConusAI-Template-1.0
// The OCR adapter against the fake platform of @conusai-kit/testing, through the real
// `@conusai-sdk/ocr` client (kit-platform's `DocumentReader`): the job is submitted, polled and read
// over HTTP exactly as the platform's async job API does it, so what is proven here is the adapter's
// calls and not a stub of them.

import { readFileSync } from "node:fs"
import { OcrJobFailed, sdkDocumentOcr } from "@app/modules/documents"
import { OcrError } from "@conusai-sdk/ocr"
import { afterEach, describe, expect, it } from "vitest"
import { closeApps, makeDocumentsApp, uploadAndProcess } from "./support/documents-app.js"
import { startOcr } from "./support/ocr.js"

const fixture = JSON.parse(
  readFileSync(new URL("../modules/documents/__tests__/fixtures/ocr-document.json", import.meta.url), "utf8"),
) as Record<string, unknown>

const open: Array<{ close(): Promise<void> }> = []
afterEach(async () => {
  await closeApps()
  await Promise.all(open.splice(0).map((platform) => platform.close()))
})

async function ocr() {
  const started = await startOcr(fixture)
  open.push(started.platform)
  return started
}

const stateOf = async (platform: { url: string }) =>
  (await (await fetch(`${platform.url}/__e2e/state`)).json()) as { ocr: unknown[] }

describe("the OCR adapter on the fake platform", () => {
  it("submits a document, polls its job and returns the document the platform read", async () => {
    const { ocr: port } = await ocr()
    const { jobId } = await port.submit(new TextEncoder().encode("%PDF-1.4"), {
      filename: "a.pdf",
      idempotencyKey: "k-1",
    })
    const read = await port.wait(jobId)
    expect(read.engine.name).toBe("fixture-engine")
    expect(read.pages).toHaveLength(3)
  })

  it("answers a repeated idempotency key with the same job, so a retried job pays once", async () => {
    const { ocr: port, platform } = await ocr()
    const bytes = new TextEncoder().encode("%PDF-1.4 same")
    const first = await port.submit(bytes, { filename: "a.pdf", idempotencyKey: "k-same" })
    const second = await port.submit(bytes, { filename: "a.pdf", idempotencyKey: "k-same" })
    expect(second.jobId).toBe(first.jobId)
    expect((await stateOf(platform)).ocr).toHaveLength(1)
  })

  it("reads a job the platform failed as permanent", async () => {
    const { ocr: port, platform } = await ocr()
    platform.ocr.failJobs(true)
    const { jobId } = await port.submit(new TextEncoder().encode("%PDF-1.4"), {
      filename: "a.pdf",
      idempotencyKey: "k-f",
    })
    await expect(port.wait(jobId)).rejects.toBeInstanceOf(OcrJobFailed)
  })

  it("draws a page of a finished job through the platform's page route: bytes, a media type, the asked dpi", async () => {
    const { ocr: port, platform } = await ocr()
    platform.ocr.setPageImage(new Uint8Array([0x89, 0x50, 0x4e, 0x47, 7]))
    const { jobId } = await port.submit(new TextEncoder().encode("%PDF-1.4"), {
      filename: "a.pdf",
      idempotencyKey: "k-p",
    })
    await port.wait(jobId)
    const image = await port.page(jobId, 2)
    expect(image?.mediaType).toBe("image/png")
    expect([...(image?.bytes ?? [])]).toEqual([0x89, 0x50, 0x4e, 0x47, 7])
    const state = (await (await fetch(`${platform.url}/__e2e/state`)).json()) as {
      ocrPages: Array<{ job: number; page: number; dpi: number }>
    }
    expect(state.ocrPages).toEqual([{ job: jobId, page: 2, dpi: 150 }])
  })

  it("reads a page the platform has no image of (a job it no longer holds, a page past the end) as no image", async () => {
    const { ocr: port } = await ocr()
    expect(await port.page(999, 1)).toBeUndefined()
    const { jobId } = await port.submit(new TextEncoder().encode("%PDF-1.4"), {
      filename: "a.pdf",
      idempotencyKey: "k-q",
    })
    await port.wait(jobId)
    // the fixture has three pages
    expect(await port.page(jobId, 4)).toBeUndefined()
  })

  it("reads a failure of the page route that is not the page's as retryable", async () => {
    const down = new OcrError("OCR request failed: network", { type: "network" })
    const adapter = sdkDocumentOcr({
      submit: async () => ({ id: 1 }) as never,
      wait: async () => ({}) as never,
      page: async () => Promise.reject(down),
    })
    await expect(adapter.page(1, 1)).rejects.toBe(down)
  })

  it("reads anything else as retryable: the error is thrown on, unchanged", async () => {
    const transient = new OcrError("OCR request failed: network", { type: "network" })
    const adapter = sdkDocumentOcr({
      submit: async () => ({ id: 1 }) as never,
      wait: async () => Promise.reject(transient),
      page: async () => ({ bytes: new Uint8Array(), contentType: "image/png" }),
    })
    await expect(adapter.wait(1)).rejects.toBe(transient)
  })

  it("carries a whole upload through the worker over HTTP: the fixture becomes proposals", async () => {
    const app = await makeDocumentsApp({ document: fixture })
    const project = await app.seedProject("acme", "owner@acme.test")
    const file = await uploadAndProcess(app, project)
    const view = (await project.call.callByName("documents.document.get", {
      ...project.slugs,
      fileId: file.fileId,
    })) as {
      document: { status: string; pending: number }
    }
    expect(view.document).toMatchObject({ status: "extracted", pending: 4 })
    expect((await stateOf(app.platform)).ocr).toHaveLength(1)
  })

  it("marks a document extracted with nothing to review when the platform reads a page with no lines", async () => {
    const app = await makeDocumentsApp({
      document: {
        ...fixture,
        page_count: 1,
        pages: [{ index: 0, width: 10, height: 10, class: "empty", text: "", lines: [], blocks: [] }],
      },
    })
    const project = await app.seedProject("acme", "owner@acme.test")
    const file = await uploadAndProcess(app, project)
    const view = (await project.call.callByName("documents.document.get", {
      ...project.slugs,
      fileId: file.fileId,
    })) as {
      document: { status: string; pending: number }
    }
    expect(view.document).toMatchObject({ status: "extracted", pending: 0 })
  })
})