fix(invoice-inbox): switch from pdfjs-dist to unpdf (#409)

* fix(invoice-inbox): switch from pdfjs-dist to unpdf for PDF text extraction After three rounds of fighting pdfjs on Vercel (#407 stubbed DOM globals, #408 tried to ship the worker file via outputFileTracingIncludes), text extraction still failed in prod with "Setting up fake worker failed" — Next's tracer can't reliably include pdfjs-dist's worker file when the package is marked as a server external. unpdf is a serverless-first wrapper around pdfjs (by unjs) that ships its own bundled pdfjs build with no canvas/worker dependencies. Drop-in replacement: extractText returns merged page text directly. - Remove DOM stubs, serverExternalPackages, outputFileTracingIncludes - Replace pdfjs-dist with unpdf (no transitive deps) - Update test mock from getDocument → extractText All 45 invoice-inbox unit tests pass. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(invoice-inbox): make unpdf import static, rename pdfjs test labels Per Greptile review on #409: 1. CRITICAL: tryExtractPdfText was still using await import('unpdf'), a dynamic import. CLAUDE.md forbids dynamic imports in extensions precisely because Next.js bundling can't reliably trace them — which is the same class of failure that caused the pdfjs prod bug. unpdf bundles statically (no canvas/worker), so a top-level static import is safe and correct. 2. NIT: two test descriptions still said "pdfjs" after the mock rename. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-07 11:32:50 +02:00
parent 3a74f0ab61
commit 8676f10fdd
5 changed files with 43 additions and 1881 deletions
@@ -1,27 +1,16 @@
 import { describe, it, expect, vi, beforeEach } from 'vitest'
 import { extractInvoiceFields } from '@/extensions/general/invoice-inbox/lib/extract-invoice-fields'

-// Mock pdfjs-dist so we can drive the regex extractors with canned text
+// Mock unpdf so we can drive the regex extractors with canned text
 // without building actual PDF binaries.
-const mockGetDocument = vi.fn()
+const mockExtractText = vi.fn()

-vi.mock('pdfjs-dist/legacy/build/pdf.mjs', () => ({
-  getDocument: (...args: unknown[]) => mockGetDocument(...args),
+vi.mock('unpdf', () => ({
+  extractText: (...args: unknown[]) => mockExtractText(...args),
 }))

 function fakePdf(text: string) {
-  return {
-    promise: Promise.resolve({
-      numPages: 1,
-      getPage: () =>
-        Promise.resolve({
-          getTextContent: () =>
-            Promise.resolve({
-              items: text.split(/\s+/).map((str) => ({ str })),
-            }),
-        }),
-    }),
-  }
+  return Promise.resolve({ totalPages: 1, text })
 }

 describe('extractInvoiceFields', () => {
@@ -40,8 +29,8 @@ describe('extractInvoiceFields', () => {
    expect(data.supplier.orgNumber).toBeNull()
  })

-  it('returns empty result when pdfjs extracts no text (image-only PDF)', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf(''))
+  it('returns empty result when unpdf extracts no text (image-only PDF)', async () => {
+    mockExtractText.mockReturnValueOnce(fakePdf(''))
    const { data, rawText } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -53,7 +42,7 @@ describe('extractInvoiceFields', () => {

  it('extracts a Luhn-valid org number', async () => {
    // 5560125790 is a valid Swedish AB org-nr (Luhn-checked)
-    mockGetDocument.mockReturnValueOnce(fakePdf('Lev: Acme AB Org.nr 556012-5790 Faktura'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Lev: Acme AB Org.nr 556012-5790 Faktura'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -63,7 +52,7 @@ describe('extractInvoiceFields', () => {
  })

  it('rejects org-nrs with bad Luhn digit', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf('Org.nr 556012-5791'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Org.nr 556012-5791'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -73,7 +62,7 @@ describe('extractInvoiceFields', () => {
  })

  it('extracts a Luhn-valid OCR reference', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf('OCR-nummer: 12345674'))
+    mockExtractText.mockReturnValueOnce(fakePdf('OCR-nummer: 12345674'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -84,7 +73,7 @@ describe('extractInvoiceFields', () => {

  it('extracts a Luhn-valid bankgiro', async () => {
    // 991-2346 is the canonical test bankgiro (Luhn-valid) used in lib/bankgiro/__tests__
-    mockGetDocument.mockReturnValueOnce(fakePdf('Bankgiro 991-2346 Plusgiro'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Bankgiro 991-2346 Plusgiro'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -94,7 +83,7 @@ describe('extractInvoiceFields', () => {
  })

  it('parses Swedish-formatted totals', async () => {
-    mockGetDocument.mockReturnValueOnce(
+    mockExtractText.mockReturnValueOnce(
      fakePdf('Att betala 12 345,67 kr')
    )
    const { data } = await extractInvoiceFields({
@@ -106,7 +95,7 @@ describe('extractInvoiceFields', () => {
  })

  it('parses Förfallodatum and normalizes to ISO', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf('Förfallodatum 2026-06-15'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Förfallodatum 2026-06-15'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -116,7 +105,7 @@ describe('extractInvoiceFields', () => {
  })

  it('extracts an invoice number after Fakturanr', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf('Fakturanr F-2024-001 Datum'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Fakturanr F-2024-001 Datum'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -126,7 +115,7 @@ describe('extractInvoiceFields', () => {
  })

  it('keeps SEK as default currency when no foreign code is present', async () => {
-    mockGetDocument.mockReturnValueOnce(fakePdf('Total 100 kr'))
+    mockExtractText.mockReturnValueOnce(fakePdf('Total 100 kr'))
    const { data } = await extractInvoiceFields({
      buffer: Buffer.from('%PDF'),
      mimeType: 'application/pdf',
@@ -135,8 +124,8 @@ describe('extractInvoiceFields', () => {
    expect(data.invoice.currency).toBe('SEK')
  })

-  it('returns empty result when pdfjs throws', async () => {
-    mockGetDocument.mockImplementationOnce(() => {
+  it('returns empty result when unpdf throws', async () => {
+    mockExtractText.mockImplementationOnce(() => {
      throw new Error('boom')
    })
    const { data, rawText } = await extractInvoiceFields({
@@ -1,10 +1,10 @@
 // Deterministic Swedish invoice field extraction.
 //
 // Replaces the deleted AI classifier. We pull text out of the PDF with
-// pdfjs-dist and run regex extractors against it. Each extractor is
-// independent — a missing field stays null rather than dragging down a
-// neighbour. Validators (Luhn for org-nr/OCR/bankgiro) keep false
-// positives near zero.
+// unpdf (a serverless-friendly pdfjs wrapper) and run regex extractors
+// against it. Each extractor is independent — a missing field stays null
+// rather than dragging down a neighbour. Validators (Luhn for
+// org-nr/OCR/bankgiro) keep false positives near zero.
 //
 // Image-only PDFs and non-PDF mime types come back with all fields null.
 // The inbox item is still created so the user can register manually.
@@ -12,21 +12,11 @@
 import type { InvoiceExtractionResult } from '@/types'
 import { normalizeOrgNumber } from '@/lib/company-lookup/normalize-org-number'
 import { validateOcrReference, validateBankgiroNumber } from '@/lib/bankgiro/luhn'
-
-// pdfjs-dist references DOMMatrix/ImageData/Path2D at module load. On
-// Vercel's Node runtime these globals don't exist; without stubs the
-// dynamic import throws "DOMMatrix is not defined" before getDocument()
-// runs. We only call getTextContent (no rendering), so empty-class stubs
-// are enough — pdfjs never invokes any methods on them.
-{
-  const g = globalThis as unknown as { DOMMatrix?: unknown; ImageData?: unknown; Path2D?: unknown }
-  if (typeof g.DOMMatrix === 'undefined') g.DOMMatrix = class {}
-  if (typeof g.ImageData === 'undefined') g.ImageData = class {}
-  if (typeof g.Path2D === 'undefined') g.Path2D = class {}
-}
+import { extractText } from 'unpdf'

 // Below this we treat the document as image-only / unreadable and skip
-// regex extraction. pdfjs-dist returns near-zero text for scanned PDFs.
+// regex extraction. The PDF text extractor returns near-zero text for
+// scanned PDFs.
 const MIN_TEXT_CHARS_FOR_EXTRACTION = 10

 export interface ExtractionInput {
@@ -108,26 +98,10 @@ async function tryExtractPdfText(input: ExtractionInput): Promise<string | null>
  if (input.mimeType !== 'application/pdf') return null

  try {
-    const pdfjs = await import('pdfjs-dist/legacy/build/pdf.mjs')
-    const loadingTask = pdfjs.getDocument({
-      data: new Uint8Array(input.buffer),
-      isEvalSupported: false,
-      disableFontFace: true,
-    })
-    const pdf = await loadingTask.promise
-
-    const pages: string[] = []
-    for (let i = 1; i <= pdf.numPages; i++) {
-      const page = await pdf.getPage(i)
-      const content = await page.getTextContent()
-      const pageText = content.items
-        .map((item) => ('str' in item ? item.str : ''))
-        .join(' ')
-      pages.push(pageText)
-    }
-    return pages.join('\n').replace(/[ \t]+/g, ' ').trim()
+    const { text } = await extractText(new Uint8Array(input.buffer), { mergePages: true })
+    return text.replace(/[ \t]+/g, ' ').trim()
  } catch (err) {
-    console.warn('[invoice-inbox/extract] pdfjs failed:', err instanceof Error ? err.message : err)
+    console.warn('[invoice-inbox/extract] pdf text extraction failed:', err instanceof Error ? err.message : err)
    return null
  }
 }
@@ -20,19 +20,6 @@ const cspDirectives = [

 const nextConfig: NextConfig = {
  output: 'standalone',
-  // Keep pdfjs-dist out of the bundle. The legacy build references
-  // @napi-rs/canvas at module load and breaks Vercel's bundling step.
-  // Text extraction works in pure Node once DOM globals are stubbed at
-  // module load (see extensions/general/invoice-inbox/lib/extract-invoice-fields.ts).
-  serverExternalPackages: ['pdfjs-dist'],
-  // Force the pdfjs worker file into the function bundle. Next.js's tracer
-  // can't see it (loaded dynamically by name), so without this it's missing
-  // in /var/task and getDocument() fails with "Setting up fake worker failed".
-  // Scoped to the invoice-inbox sub-path — other extensions don't use pdfjs
-  // and shouldn't pay the ~1 MB worker cost.
-  outputFileTracingIncludes: {
-    '/api/extensions/ext/invoice-inbox/**': ['./node_modules/pdfjs-dist/legacy/build/pdf.worker.mjs'],
-  },
  async redirects() {
    return [
      {
@@ -45,7 +45,6 @@
    "lucide-react": "^0.563.0",
    "next": "16.1.5",
    "next-themes": "^0.4.6",
-    "pdfjs-dist": "^5.4.530",
    "qrcode": "^1.5.4",
    "react": "19.2.3",
    "react-dom": "19.2.3",
@@ -57,6 +56,7 @@
    "sharp": "^0.34.5",
    "svix": "^1.85.0",
    "tailwind-merge": "^3.4.0",
+    "unpdf": "^1.6.2",
    "web-push": "^3.6.7",
    "xlsx": "^0.18.5",
    "zod": "^4.3.6"